CN101504760A - Digital image concealed information detecting and positioning method - Google Patents

Abstract

The invention relates to a method for detecting and positioning hidden information in a digital image, which comprises the following steps: detecting a digital image containing hidden information by using a blocking effect computing method; if the digital image is in a JPEG image compressed format, extracting a quantization matrix of the JPEG; deducing a quantization DCT coefficient of an original image before the embedment of the hidden information according to a DCT coefficient of the digital image; and restoring the pixel value of the original image according to the DCT coefficient of the original image, determining the embedment position of the hidden information according to the difference between the pixel values of the original image and the digital image, and estimating the embedding quantity of the hidden information. The technical proposal of the invention overcomes the drawback that the prior hidden information detection technology cannot accurately position the embedment position of the hidden information, and can detect and position hidden information hidden by various steganography algorithms in the digital image.

Description

A kind of digital image concealed information detects the method with the location

Technical field

The present invention relates to information security field, relate in particular to a kind of method that in digital picture, detects with positioning hidden information.

Background technology

Information hiding is hidden in confidential information does not in fact exactly reveal weak point among the common information.Carrier difference according to hiding Info can be divided into the Information hiding in image, video, sound, text etc., and the information of being hidden also can be above various forms, just all they is considered as bit stream when hiding and handles.The ultimate principle of Information hiding is to have utilized the insensitivity of human sensorium to some details, carrier is done some small variations, and do not cause observer's suspection.For different carriers different characteristics are arranged again, utilized human vision property as image and video information hiding, and different to the insensitivity of rest image and video vision.The hiding auditory properties that utilizes people's ear of voice signal, therefore, the Information hiding in different carriers has its different characteristics.

The opposite of Information hiding is exactly that information detects, and the meaning that information detects research is, can impel Information Hiding Techniques by legal use on the one hand, can further promote the further investigation of hidden algorithm on the other hand.Although Information hiding can not cause human attention on sense organ, but still can find the variation of carrier from the data analysis, the problem that the context of detection that hides Info need be studied is a lot, such as analyzing caused feature such as the change of statistical property in carrier of various hidden algorithms, the change of palette, the change of local edge, change of eigenwert or the like.Analyze can the hide Info upper limit of capacity of various carriers, set up and the corresponding to objective evaluation criterion of subjective assessment criterion.In addition, the foundation that information detects research framework also is a very important problem, sets up and the similar analyzing detecting method of cryptanalysis, as only knowing the detection of latent literary composition, known carrier detection, known message detection, algorithm known detection etc.

Information Hiding Techniques is an importance of information security field, and the analytical technology of Information hiding has become the focus of information security field research as the opposite of Information Hiding Techniques.Image has obtained application in many-side as the good carrier of Information hiding.Though proposed algorithm and systems that a plurality of Information hiding based on image detect, all there are benefit and limitation separately in every kind of algorithm or system.So far still also there is not to form theory based on the Information hiding detection technique of image.Therefore, be still waiting to further develop based on the Information hiding detection technique of image and perfect.

Extract the final goal that concealed information is latent close analytical technology from latent close image, the extraction of concealed information comprises the embedded quantity of estimating concealed information and the embedded location of concealed information.Though the realization difficulty of this process is very big, the plurality of detection algorithm has been arranged when whether detection exists concealed information, can more accurately estimate the embedded quantity of secret information.Up to the present still there is not accurately to determine the report of the embedded location of concealed information.

Summary of the invention

The objective of the invention is a kind of digital image concealed information and detect and the method for locating, more particularly, this method can not only detect whether contain concealed information in the digital picture, and can also detect the embedded location of concealed information.

Technical scheme of the present invention is that a kind of digital image concealed information detects and localization method, it is characterized in that, may further comprise the steps:

Detect the described digital picture that has comprised concealed information with blocking effect computing method;

If described digital picture is a jpeg image compressed format, then extract the quantization matrix of JPEG;

Derive the quantization DCT coefficient that embeds the original image before the concealed information according to described digital picture DCT coefficient; And

According to the DCT coefficient of described original image, recover the pixel value of this original image, judge the embedded location of described concealed information by the difference of this original image and described Pixel of Digital Image value, estimate the embedded quantity of this concealed information.

Describedly may further comprise the steps with blocking effect computing method:

If described data image signal be x (m, n) | m ∈ [1, M], n ∈ [1, N] }, the level error of image is divided into:

d _h(m，n)＝x(m，n+1)-x(m，n)，n∈[1，N-1]，

Come the size of quantize block effect with the mean value of block boundary place pixel difference, calculating formula is as follows:

$B_h=\frac{1}{M(K-1)}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{K-1}{\mathrm{\Σ}}}\left|d_h(i,8j)\right|$

$A_h=\frac{1}{M(N-1)}\underset{i}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N-1}{\mathrm{\Σ}}}\left|d_h(i,j)\right|-\frac{1}{M(K-1)}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{K-1}{\mathrm{\Σ}}}\left|d_h(i,8j)\right|$

Wherein, $K=\left[\frac{N}{8}\right]-1,$ B _hBe the interblock pixel value difference and, A _hBe piece interior pixel difference and;

Make the vertical difference of described digital picture be divided into:

d _v(m，n)＝x(m+1，n)-x(m，n)，m∈[1，M-1]，

Obtain B _vAnd A _v:

$B_v=\frac{1}{N(K-1)}\underset{i=1}{\overset{K-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\left|d_v(8i,j)\right|$

$A_v=\frac{1}{M(M-1)}\underset{i}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\left|d_v(i,j)\right|-\frac{1}{N(K-1)}\underset{i=1}{\overset{K-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\left|d_v(8i,j)\right|$

Wherein, $K=\left[\frac{M}{8}\right]-1,$ B _vBe the interblock pixel value difference and, A _vBe piece interior pixel difference and;

Obtain level and vertical direction piece interior pixel difference and be:

A＝A _h+A _v

Level and vertical direction interblock pixel value difference and be:

B＝B _h+B _v；

If the absolute value of the difference of B and A judges that greater than 0.3 described digital picture is a jpeg image compressed format.

When deriving the quantization DCT coefficient of the original image before embedding concealed information according to described digital picture DCT coefficient, with the quantization DCT coefficient of described digital picture quantization DCT coefficient as described original image.

The beneficial effect of technical scheme of the present invention is to have solved the shortcoming that existing concealed information detection technique can not accurately be located the embedded location that hides Info, and can multiplely latent write concealed information that algorithm hides and detects and locate utilizing in the digital picture.

Description of drawings

Fig. 1 is the JPEG test pattern that does not embed concealed information in one embodiment of the invention

Fig. 2 is the digital picture that embeds concealed information in one embodiment of the invention

Fig. 3 is concealed information embedded location figure in one embodiment of the invention

Fig. 4 is the graph of a relation of the size and the quality factor of jpeg image blocking effect in one embodiment of the invention

Fig. 5 tries to achieve JPEG quality factor EQ curve map in one embodiment of the invention

Fig. 6 is the concealed information embedded location figure that estimates from latent close digital picture in one embodiment of the invention

Fig. 7 is that digital image concealed information detects and the localization method process flow diagram in one embodiment of the invention

Embodiment

Further specify technical scheme of the present invention below in conjunction with the drawings and specific embodiments.

As shown in Figure 7, a kind of digital image concealed information provided by the invention detects and localization method, comprises following steps:

S101, detect latent close digital picture with blocking effect computing method and whether compressed with jpeg format;

, extracts this image of S102 the quantization matrix of JPEG if being compressed by JPEG;

S103, the latent close image DCT coefficient of basis are derived the quantization DCT coefficient that embeds the preceding original image of concealed information;

S104, according to the DCT coefficient of original image, recover the pixel value of original image, judge the embedded location of concealed information by the original image and the difference of latent close image pixel value, and estimate the embedded quantity of concealed information thus.

Because many information disguising softwares are as EZStego, Hide﹠amp; Seek, S-Tools4, Steganos, StegoDos etc., the method that has all adopted Spatial LSB to replace embeds information, contains close image with non-compression storage format storage mostly after the embedding information.Detection to this class image can not an extension name according to image file.To carry out quantitative test to image pixel value, whether be compressed in the past by JPEG with the image of finding non-compression storage format.Be the image that does not embed concealed information as shown in Figure 1, picture format is Jpeg, and the image size is 256 * 256.Fig. 2 is the image of using S-Tools4 embeds concealed information in Fig. 1 after, and picture format is bmp.Expression for convenience, the embedded location of concealed information is represented with bianry image, as shown in Figure 3, size is 256 * 256, every embeds the 1bit data, and the figure intermediate value is to represent that this position embeds information 1 (deceiving), and value is to represent that this position does not embed information 0 (in vain), latent close image is embedded in the position that the bianry image intermediate value is 1 (deceiving), and the bianry image intermediate value is that the position of 0 (in vain) does not embed information.

JPEG is based on the lossy compression of DCT piece, and the DCT coefficient of each 8x8 piece is quantized, rounds the loss that causes the DCT coefficient.Quantization operation is brought blocking effect to image.Blocking effect is owing to the block boundary place, and the uncontinuity that signal changes causes, and is that unit carries out because quantize with the piece.Obviously, the ratio of compression of JPEG is high more, and blocking effect is just obvious more.

The ultimate principle of jpeg image blocking effect computing method is:

If picture signal be x (m, n) | m ∈ [1, M], n ∈ [1, N] }, the level error of image is divided into:

d _h(m，n)＝x(m，n+1)-x(m，n)，n∈[1，N-1]

Come the size of quantize block effect with the mean value of block boundary place pixel difference, computing method are as follows:

$B_h=\frac{1}{M(K-1)}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{K-1}{\mathrm{\Σ}}}\left|d_h(i,8j)\right|$

(1)

$A_h=\frac{1}{M(N-1)}\underset{i}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N-1}{\mathrm{\Σ}}}\left|d_h(i,j)\right|-\frac{1}{M(K-1)}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{K-1}{\mathrm{\Σ}}}\left|d_h(i,8j)\right|$

(2)

Wherein, $K=\left[\frac{N}{8}\right]-1.$ B _hBe the interblock pixel value difference and, A _hBe piece interior pixel difference and.Use the same method and to calculate the B of vertical direction _vAnd A _vThe vertical difference of image is divided into:

d _v(m，n)＝x(m+1，n)-x(m，n)，m∈[1，M-1]

B _vAnd A _vComputing method as follows:

$B_v=\frac{1}{N(K-1)}\underset{i=1}{\overset{K-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\left|d_v(8i,j)\right|$

(3)

$A_v=\frac{1}{M(M-1)}\underset{i}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\left|d_v(i,j)\right|-\frac{1}{N(K-1)}\underset{i=1}{\overset{K-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\left|d_v(8i,j)\right|$

(4)

Wherein, $K=\left[\frac{M}{8}\right]-1.$ B _vBe the interblock pixel value difference and, A _vBe piece interior pixel difference and.

Consideration level and vertical direction piece interior pixel difference and be:

A＝A _h+A _v (5)

Level and vertical direction interblock pixel value difference and be:

B＝B _h+B _v (6)

Detect principle according to top JPEG blocking effect, calculate level and vertical direction piece interior pixel difference and A, level and vertical direction interblock pixel value difference and B at Fig. 2.The size of jpeg image blocking effect is decided by the difference of B and A, if piece image was not compressed by JPEG, the absolute value of the difference of B and A is very little, generally is no more than 0.3, and after image was by the JPEG compression, the difference of B and A can increase.Detect effect as shown in Figure 4.

Extract the method for JPEG quantization matrix, comprise following steps:

If image pixel be x (m, n) | m ∈ [1, M], n ∈ [1, N] }, at first image is divided into some 8x8 sizes piece B (i, j), wherein $1\≤i\≤\left[\frac{M}{8}\right],$ $1\≤j\≤\left[\frac{N}{8}\right].$ Each 8x8 piece is carried out dct transform, rounds then and obtain:

Y ₁(i，j)＝DCT(B(i，j))

Y ₂(i，j)＝[Y1(i，j)]

When quality factor q satisfies: during 1≤q≤50, calculate following formula:

$\mathrm{EQ}\left(q\right)=\underset{i,j}{\mathrm{\Σ}}|Y_2(i,j)-\left(\frac{Q*50}{q}\right)*[Y_2(i,j)*\frac{q}{(Q*50)}]|---\left(7\right)$

When quality factor q satisfies: during 51≤q≤99, calculate following formula:

$\mathrm{EQ}\left(q\right)=\underset{i,j}{\mathrm{\Σ}}|Y_2(i,j)-Q*(2-\frac{q}{50})*\left[\frac{Y_2(i,j)}{Q*(2-\frac{q}{50})}\right]|---\left(8\right)$

Q is the standard quantization table in the formula, the EQ curve that draws, if image was compressed by JPEG, the local minimum on the EQ curve is exactly the quality factor of JPEG, just can calculate the quantization matrix of JPEG according to quality factor.

According to (7) and (8) formula EQ curve that draws, as shown in Figure 5, if image was compressed by JPEG, the local minimum on the EQ curve is exactly the quality factor of JPEG, just can calculate the quantization matrix of JPEG according to quality factor at Fig. 2.

And the localization method of the estimation of quantization DCT coefficient and concealed information then comprises following steps:

If carrier image is { I _c(m, n) | m ∈ [1, M], n ∈ [1, N] }, after the spatial domain hides concealed information with the LSB method, obtain containing close image { I _s(m, n) | m ∈ [1, M], n ∈ [1, N] }, I _sWith I _cFollowing relation is arranged:

I _s(m，n)＝I _c(m，n)+W(m，n) (9)

Because the LSB hidden method only changes the lowest order of pixel, the influence of each pixel value is equivalent to add one+1 ,-1 or 0.Therefore, (m, n) being equivalent to one, to have only value be+1 to W ,-1 or 0 matrix.By the embedding of LSB rule decision W (m, n) intermediate value is 0 probability P (0)=0.5, value is 1 probability P (1)=0.25, is worth the probability P (1)=0.25 for-1.Because I _c(m n) is the image of jpeg format, and the problem that will solve is how from I below _s(m recovers I in n) _c(m, quantization DCT coefficient n).

Dct transform uses following formula to calculate:

$F(u,v)=\frac{1}{4}C\left(u\right)C\left(v\right)\left[\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}f(i,j)\cos \frac{(2i+1)\mathrm{u\π}}{16}\cos \frac{(2j+1)\mathrm{v\π}}{16}\right]$

Its inverse transformation uses following formula to calculate:

$f(i,j)=\frac{1}{4}C\left(u\right)C\left(v\right)\left[\underset{u=0}{\overset{7}{\mathrm{\Σ}}}\underset{v=0}{\overset{7}{\mathrm{\Σ}}}F(u,v)\cos \frac{(2i+1)\mathrm{u\π}}{16}\cos \frac{(2j+1)\mathrm{v\π}}{16}\right]$

Above in two formulas,

$C\left(u\right),C\left(v\right)=\frac{1}{\sqrt{2}},$ Work as u, v=0

C (u), C (v)=1, other

To I _s(m n) carries out dct transform, obtains DC coefficient to be:

$F_s\left(\mathrm{0,0}\right)=\frac{1}{4}C\left(0\right)C\left(0\right)\left[\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}{I}_s(i,j)\right]=\frac{1}{8}\left[\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}(I_c(i,j)+W(i,j))\right]$

$=F_c\left(\mathrm{0,0}\right)+\frac{1}{8}\left[\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}W(i,j)\right]---\left(10\right)$

Obtaining ac coefficient is:

$F_s(u,v)=\frac{1}{4}C\left(u\right)C\left(v\right)\left[\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}{I}_s(i,j)\cos \frac{(2i+1)\mathrm{u\π}}{16}\cos \frac{(2j+1)\mathrm{v\π}}{16}\right]$

$=\frac{1}{4}\left[\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}(I_c(i,j)+W(i,j))\cos \frac{(2i+1)\mathrm{u\π}}{16}\cos \frac{(2j+1)\mathrm{v\π}}{16}\right]$

$=F_c(u,v)+\frac{1}{4}\left[\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}W(i,j)\cos \frac{(2i+1)\mathrm{u\π}}{16}\cos \frac{(2j+1)\mathrm{v\π}}{16}\right]$

$\≈F_c(u,v)+\frac{1}{4}\left[\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}W(i,j)\right]$

(11)

Owing to known carrier image I _c(m, quantization matrix Q n) can be with Q to containing close image I _s(m, DCT coefficient n) further quantize to obtain:

$F_s^Q\left(\mathrm{0,0}\right)\≈(F_c\left(\mathrm{0,0}\right)+\frac{1}{8}\left[\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}W(i,j)\right])/Q\left(\mathrm{0,0}\right)$

(12)

With

$F_s^Q(u,v)\≈(F_c(u,v)+\frac{1}{4}\left[\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}W(i,j)\right])/Q(u,v)$

(13)

Be the quantization DCT coefficient of carrier image,

Be the quantization DCT coefficient that contains close image, the difference between them with

Size relevant.By preceding surface analysis as can be known, W (i, value j) has only+and 1 ,-1 and 0, and+1 and-1 probability approximately equal that occurs, so formula

Absolute value very little.Therefore, can reach a conclusion, when the quantification step-length is bigger,

With

Substantially equal, when the quantification step-length is smaller,

With

A small amount of error is arranged.

Can be by formula (10)～(11) only according to containing the quantization DCT coefficient that close image is derived original image (carrier image), to contain the quantization DCT coefficient of the quantization DCT coefficient of close digital picture as this original image, DCT coefficient according to original image, can recover the pixel value of original image, just can judge the embedded location of concealed information by original image and the difference that contains close image pixel value, thereby estimate the embedded quantity of concealed information.As shown in Figure 6,0 this position of expression does not embed information, and 1 this position of expression embeds information.Can estimate the embedded quantity of concealed information according to Fig. 6.

Confirm by experiment, when the quality factor of original image is 50 or 50 when following,,, when being 75, the quality factor of original image can reach more than 92% up to more than 98% the correct recognition rata of concealed information embedded location the correct recognition rata of concealed information embedded location.

Claims (3)

1, a kind of digital image concealed information detects and localization method, it is characterized in that, may further comprise the steps:

Detect the described digital picture that has comprised concealed information with blocking effect computing method;

If described digital picture is a jpeg image compressed format, then extract the quantization matrix of JPEG;

Derive the quantization DCT coefficient that embeds the original image before the concealed information according to described digital picture DCT coefficient; And

According to the DCT coefficient of described original image, recover the pixel value of this original image, judge the embedded location of described concealed information by the difference of this original image and described Pixel of Digital Image value, estimate the embedded quantity of this concealed information.

2, digital image concealed information as claimed in claim 1 detects and localization method, it is characterized in that, describedly may further comprise the steps with blocking effect computing method:

If described data image signal be x (m, n) | m ∈ [1, M], n ∈ [1, N] }, the level error of image is divided into:

d _h(m，n)＝x(m，n+1)-x(m，n)，n∈[1，N-1]，

Come the size of quantize block effect with the mean value of block boundary place pixel difference, calculating formula is as follows:

$B_h=\frac{1}{M(K-1)}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{K-1}{\mathrm{\Σ}}}\left|d_h(i,8j)\right|$

$A_h=\frac{1}{M(N-1)}\underset{i}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N-1}{\mathrm{\Σ}}}\left|d_h(i,j)\right|-\frac{1}{M(K-1)}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{K-1}{\mathrm{\Σ}}}\left|d_h(i,8j)\right|$

Wherein, $K=\left[\frac{N}{8}\right]-1,$ B _hBe the interblock pixel value difference and, A _hBe piece interior pixel difference and;

Make the vertical difference of described digital picture be divided into:

d _v(m，n)＝x(m+1，n)-x(m，n)，m∈[1，M-1]，

Obtain B _vAnd A _v:

$B_v=\frac{1}{N(K-1)}\underset{i=1}{\overset{K-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\left|d_v(8i,j)\right|$

$A_v=\frac{1}{M(M-1)}\underset{i}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\left|d_v(i,j)\right|-\frac{1}{N(K-1)}\underset{i=1}{\overset{K-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\left|d_v(8i,j)\right|$

Wherein, $K=\left[\frac{M}{8}\right]-1,$ B _vBe the interblock pixel value difference and, A _vBe piece interior pixel difference and; Obtain level and vertical direction piece interior pixel difference and be:

A＝A _h+A _v

Level and vertical direction interblock pixel value difference and be:

B＝B _h+B _v；

If the absolute value of the difference of B and A judges that greater than 0.3 described digital picture is a jpeg image compressed format.

3, digital image concealed information as claimed in claim 1 detects and localization method, it is characterized in that, when deriving the quantization DCT coefficient of the original image before embedding concealed information according to described digital picture DCT coefficient, with the quantization DCT coefficient of described digital picture quantization DCT coefficient as described original image.

Images